The Application Of Support Vector Machine In Neuron Classification With Spatial Structure

The brain is the human’s organization which has the most complicated structure andfunction and it contains kinds of neurons. In order to study the human brain the neuronmust be studied which comes from the brain, and in order to know neuron it isnecessary to know what the neuron is and how the neuron is classified. It is a veryimportant task that quickly and accurately classifying neuron. This task can makepeople easy to know the neuron. It is very meaningful and worthwhile to do this task.The main research of this paper is neuron and the entrance of research is the spatialstructure of neuron. It is classified by the feature of the spatial structure. It needs tocombine kinds of methods in the spatial structure of neuron. At last the neuron will beclassified by Support Vector machine.The main innovations are as follows：(1) The spatial structure of the neuron has the feature of fractal geometry in somewhere.By observation and comparison the spatial structure and to use the feature of its own,it is proposed an algorithm of fractal dimension and the fractal dimension of neuronis calculated. The fractal dimension is proved to classify the neuron is a valid anduseful by experiment.(2) The data of spatial structure of neuron is from neuromorpho. It is a database whichcollects the data of various biological laboratories. Because of the data is fromvarious laboratories, the data forms the unbalanced problem. In this paper, itcombines the common way of solution the unbalanced problem and proposes afitness to solve the unbalanced problem. And then it is proved is effective byexperiment.(3) The feature selection of spatial structure of neuron. There are many features fromthe spatial structure. It is a very important problem that how to choose the bestfeather of spatial structure. In this paper, it uses the signal to noise ratio to solve thisproblem. The signal to noise ratio (SNR) is easy to understand and calculate.To useSNR can get the better feather group of spatial structure.And then there will be aexperiment to prove it is valid.(4) After the data pre-processing over. At last, it will use the SVM to classify the neuron.By chose different kernel function, it gets the best result of classifying the neurons. Then there will be an experiment to prove it.Finally, the highest classification result is86.7%.It is from the way which combinesthe various methods. And the result shows that this method is valid to classify theneuron. At the same time it gives a new way to classify the neuron.